For Heads-of · Practitioner
Performance Metric
A quantitative measure of how well an AI model or system performs a task, such as accuracy, precision, recall, F1 score or a domain-specific measure.
- performance metrics
- evaluation
- accuracy
- measurement
Common metrics
Accuracy (the share of correct predictions), precision (of the positive predictions, how many were right), recall (of the actual positives, how many were found) and F1 score (a balance of precision and recall).
Limitation
A single metric rarely tells the whole story. A model can be accurate overall while performing poorly for a minority group, discriminating despite a good aggregate figure, and the right metric depends on which errors cost most in the use case.
Governance requirement
Track metrics broken down by group, use case and time, and set thresholds in advance. Not breaking results down by group is the disparate impact monitoring gap; AI system monitoring and logging carries the metrics into production.
Related Terms
Confusion Matrix
A table comparing a classifier's predictions with the actual outcomes, from which precision, recall and false positive and false negative rates are calculated.
Model Monitoring
Continuous observation of a deployed model's inputs, outputs, performance and behaviour, to detect degradation, drift, misuse or security anomalies.
Performance Regression
A decline in model performance over time, typically caused by shifts in the data, concept drift or changes in the environment the model operates in.
Testing AI Systems
Systematic evaluation of AI models and systems to verify that they perform as intended, identify failure modes and assess risks.
Related risks and controls
AI model drift and degradation
Drift is a metric moving; without a chosen metric and a baseline it cannot be seen.
Disparate impact monitoring gap
A metric averaged over everyone can be stable while a subgroup's is not.
AI system monitoring and logging
Where the metric is actually recorded over time, without which a baseline is a memory.